I’m not sure if Retrievr has any commercial application, but it is a really beautiful piece of technology.
Retrievr has a Flash sketch pad built into the site. Draw something – anything – and it will fetch Flickr images that are similar. My very rough drawing of a black line intersecting with a red blog brought up some nice results (see image).
More information is available on their about page, here. Thanks Nivi for pointing this out to me (and I aplogize again for using your “special” towel to dry off Laguna’s feet 
).
My drawing gave me this result:
“Error in application retrievr
Maximum Thread count reached.”
Looks like Retriver has been Crunch’ed! (a la Slashdotted)
I played with this for an good hour yesterday – good stuff.
Crap. Really. Didn’t show even 1 relevant pic for an apple.
Nivi has a “special” towel? How rich! Looking forward to harassing him about that.
This is an example of CBIR, or Content Based Image Retrieval. CBIR is a very difficult problem. I like this CBIR implementation, and would love to see them take it forward to the point of being a useful tool. For an interesting approach, see blobworld.
that “special” towel wouldn’t be the one Nivi keeps under his bed would it?
The commercial application for this is as a new form of search tool. Rather than combining this with Flickr, imagine ti combined with something like Google images or Yahoo images.
No longer do you need to know the words to describe what you’re after, you just need to be able to draw it.
You’ll also turn up all those images that are labelled in a language other than English, or whatever your native language is.
And of course someone will no doubt work out how to turn it into a p0rn search tool.
The app seems to submit a new request every time you draw a single line in the Flash app. It would be considerable less draining on the server if it only did it when you click a “submit” button (and more intuitive, I think).
A cool one.
If you draw a face using skin/eyes/hair colors (really a circle with something similar to eyes, mouth and hairs), a good part of the results are faces.
MS bought a similar company a few years ago (1999?). It was founded by Dave Salesin. IIRC, they have a patent on this…
you can look some of my screenshot at http://mv.vexite.com it’s really interesting. I like this web app!
Anyone who has searched the stock image libraries for compositional matches (rather than content-based matches) will instantly see the commercial applications of this. Getty Images, Corbis–all of them should be running something like this.
retrievr is pretty impressive. Hope it makes it into production. We need new ways to search for web content.
I really think that retrievr could be a very interesting part of systemone. Image not only being able to being shown context sensitive information on your text / text documents but also on your images? What if systemone would show related images to the one you have just drawn / uploaded?
… the more I get into systemone the more I’m convinced that this app is gonna rock the world.
imageImagineI like very much the writings and pictures and explanations in your adress so I look forward to see your next writings.
To provide useful information, please click to view
Bose headphones
ghd Hair Straightener
Women is Dakota
Sundance UGG Boots
Thank you!